The decision to switch a campaign off is made badly in two directions. It gets killed on day four because a bad weekend made the numbers look frightening, or it survives for months because someone believes in it and there is always a reason this month was unusual.
Both happen because the rule was never written down. Once money is being spent, every conversation about stopping is emotional: sunk cost on one side, panic on the other, and a data set small enough to support either argument.
So the answer is not a better instinct in the moment. It is a written exit, agreed before launch, that names the number, the volume of evidence required, and the date the decision gets made. After that, turning it off is not a judgement call. It is bookkeeping.
Write the exit before you write the ad
Before a campaign launches, four things should be on the page next to it.
What it is for. Not just orders, but which of the four numbers it is meant to move and for which products. A campaign meant to buy first orders from people who have never heard of you is judged differently to one meant to catch people already searching your name.
What success looks like, as a cost per order or a return, taken from the margin ceiling for those products rather than from a general target.
How much evidence is needed before the number can be trusted. This is a count of conversions, not a number of days.
And when the decision is made. An actual date, so nobody has to raise it awkwardly.
Writing those four takes ten minutes and removes most of the later argument. It also improves the campaign itself, because a campaign with no defensible answer to the first question usually should not launch. If you cannot say what it is for beyond wanting more sales, you will not be able to say later whether it worked.
Write down the number that ends the campaign on the day you turn it on.
Enough data means conversions, not days
A week is not a sample. Neither is a month, on a store with low volume.
What makes a result readable is the number of conversions behind it. A campaign with three orders tells you almost nothing about its true cost per order, because one more order or one fewer swings it by a third. The same campaign with forty orders is telling you something real. That is why two stores can run identical tests and one has an answer in eight days while the other still does not have one after two months.
So before launching, work out roughly how long it takes to accumulate a readable number of conversions at the budget you are willing to spend. If the answer is longer than you are prepared to wait, the honest response is to change the test rather than to read the small numbers anyway. Test something with more volume behind it, widen the scope, or accept a rougher signal and say so out loud.
There is a related trap in the other direction. A campaign that has spent a lot but converted little has already told you something, and waiting for a full sample of a failure is just an expensive way to be sure. If spend has passed several times the cost per order you would accept and there are no orders, that is evidence, not an incomplete sample.
Three failures that are not the campaign's fault
Before switching anything off, check whether the campaign is being blamed for something downstream of it.
The landing page. If clicks arrive on a page that is slow, out of stock, or not the thing the ad implied, the campaign is doing its job and the store is losing the visitor. Turning it off hides the problem and it will reappear on the next campaign.
The feed or the product data. In shopping, a campaign can only work with the products that are eligible and visible. A rejected share of the catalogue, a stale price, an availability mismatch, and the campaign is competing with one hand tied. That is a data fix, not a bidding decision.
The offer. If the product is out of stock in the popular sizes, priced above where the comparison sits, or shipped on a timeline that loses to the alternative, no amount of campaign work rescues it.
The test is simple. Ask whether another campaign pointed at the same products and pages would fail in the same way. If the answer is yes, you have found a store problem, and turning off the campaign only stops you seeing it.
Pause, cut and restructure are three different moves
Off is not the only option, and reaching for it first loses information you paid for.
Cutting scope is often the right move. A campaign losing money overall frequently contains a subset that is profitable: one product group, one region, one placement, one device. Removing the losing part keeps the working part and keeps the history.
Restructuring is the move when the problem is that good and bad are averaged together inside one bucket. Splitting by margin band or by product group lets the ceiling apply where it belongs, and the same spend produces a different result without a single new ad.
Pausing is for a specific, temporary cause. Stock gone, a shipping delay, a site problem, a season that has ended. Pause with a note about what has to be true to restart, or it will sit paused forever and quietly become a decision nobody made.
Turning it off entirely is for the campaign that has had enough evidence, has no profitable subset, and cannot be fixed by structure. That is a smaller set than the number of campaigns that get switched off in a nervous week.
A worked exit rule
Invented numbers, to show the shape of a rule rather than to prescribe one.
Say a product group contributes $35 an order, so the ceiling is $35 and the target is $25 to leave room. The rule might read: review at 30 orders or $1,500 spent, whichever comes first. If cost per order is under $25, continue and consider more budget. If it is between $25 and $35, keep it running at the current budget and fix the largest downstream item on the list. If it is above $35 with no profitable subset, switch it off and record why.
Add one guard rail for the impatient direction: no decision before the review point except for something categorically wrong, like spend going somewhere obviously irrelevant or a tracking break.
And add a date. Reviewed on the fifteenth, decided the same day.
The numbers in your rule will come from your own contribution arithmetic and your own volume. What matters is that they exist in writing before the spending starts, because that is what makes the decision cheap when it arrives.
What switching it off does to everything else
A campaign is not isolated, so the decision has consequences beyond its own line in the report.
The budget goes somewhere. If it moves to a campaign that was already delivering, expect the marginal orders it buys to cost more than that campaign's current average, because the cheap demand there is already being bought. Plan for that rather than being surprised by it.
The demand may not disappear. Some of the orders a campaign was buying would have arrived anyway through search, direct or email. That is exactly why turning something off can look better in the account than it does in the bank, and why the number worth watching in the week after is total orders and blended cost, not the surviving campaign's own numbers.
And the platform loses the conversion history attached to that campaign. Automated bidding uses recent conversion data, so switching things off and on repeatedly makes every version of the campaign worse than a stable one would have been.
Which is the last argument for writing the rule in advance. The cost of turning a campaign off is not zero, so it should be done once, deliberately, and not three times a quarter on a feeling.